Sustainable Location Selection of Data Centers
Developing a Multi-Criteria Set-Covering Decision-Making Methodology
Kheybari, Siamak; Davoodi Monfared, Mansoor; Farazmand, Hadis; Rezaei, Jafar DOI10.1142/S0219622020500157
Publication date 2020
Document Version Final published version Published in
International Journal of Information Technology and Decision Making
Citation (APA)
Kheybari, S., Davoodi Monfared, M., Farazmand, H., & Rezaei, J. (2020). Sustainable Location Selection of Data Centers: Developing a Multi-Criteria Set-Covering Decision-Making Methodology. International Journal of Information Technology and Decision Making, 19(3), 741-773.
https://doi.org/10.1142/S0219622020500157 Important note
To cite this publication, please use the final published version (if applicable). Please check the document version above.
Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons. Takedown policy
Please contact us and provide details if you believe this document breaches copyrights. We will remove access to the work immediately and investigate your claim.
This work is downloaded from Delft University of Technology.
'You share, we take care!' - Taverne project
https://www.openaccess.nl/en/you-share-we-take-care
Otherwise as indicated in the copyright section: the publisher
is the copyright holder of this work and the author uses the
Dutch legislation to make this work public.
Sustainable Location Selection of Data Centers: Developing a
Multi-Criteria Set-Covering Decision-Making Methodology
Siamak Kheybari*,§, Mansoor Davoodi Monfared†,¶, Hadis Farazmand*,|| and Jafar Rezaei‡,**
*Department of Management Ferdowsi University of Mashhad
Mashhad, Iran
†Department of Computer Sciences and Information Technology
Institute for Advanced Studies in Basic Sciences Zanjan, Iran
‡Faculty of Technology, Policy and Management
Delft University of Technology Ja®alaan 5, 2628 BX Delft, The Netherlands
§Kheybari@mail.um.ac.ir; Siamak.Kheybari@gmail.com ¶mdmonfared@iasbs.ac.ir ||h92faraz@yahoo.com **J.Rezaei@tudelft.nl Received 25 October 2019 Revised 24 March 2020 Accepted 26 March 2020 Published 24 June 2020
In this paper, a multi-criteria set-covering methodology is proposed to select suitable locations for a set of data centers. First, a framework of criteria, with social, economic and environmental dimensions, is presented. The framework is used to calculate the suitability of potential data center locations in Iran. To that end, a sample of specialists in Iran was asked to take part in an online questionnaire, based on best–worst method (BWM), to determine the weight of the criteria included in the proposed framework, after which a number of potential locations are evaluated on the basis of the criteria. The proposed model is evaluated under a number of settings. Using the proposed multi-criteria set-covering model, not only the utility of candidate places is evaluated by sustainability criteria but also all service applicants are covered by at least one data center with a speci¯c coverage radius.
Keywords: Data center; location selection; best–worst method (BWM); sustainability framework; linear set-covering model; conditional performance.
1. Introduction
At the start of the third millennium, most researchers and practitioners have acknowledged the crucial role information technology plays in the economic, political §Corresponding author.
Vol. 19, No. 3 (2020) 741–773
°c World Scienti¯c Publishing Company DOI:10.1142/S0219622020500157
and cultural development of today's societies. In recent years, computer networks have evolved considerably, for instance, when it comes to data processing and storage centers. A data center consists of a set of servers, communication infrastructures and
electronic devices that are used to store, provide and support network services.1,2
The primary functions of such centers are connecting large numbers of servers at low
cost, analyzing data and indexing information in a safe space.3,4For most countries,
choosing a suitable location for their data centers is important, because (i) out-sourcing the national data and information to data centers located abroad poses a
potential threat to national security5,6; (ii) using these centers facilitates the sharing
of data and information among people and organizations and (iii) individuals and organizations can access data, information, analysis and meta-analysis more quickly. Selecting a suitable location is one of the main steps in this process. Although multi-criteria decision making (MCDM) methods are popular when it comes to se-lection of a suitable location of data centers, on their own, MCDM methods are not e®ective enough, mainly due to their inability to handle the distribution of the clients of the data centers. That is to say, there is a chance that some customers are not covered by a given data center, even if the distance between the data centers and the applicants is included as a criterion. To address this crucial de¯ciency, a multi-criteria methodology is proposed in this paper that directly takes the distance into account. Using the proposed methodology, the selected locations not only have the proper quali¯cations with respect to the decision-making criteria (used in MCDM methods), but they are also guaranteed to cover all the applicants of the data centers, which guarantee acceptable speed and support services for all.
To develop the proposed methodology, ¯rst a sustainability framework is pro-posed to identify the criteria that a®ect the location of the data center, divided into economic, social and environmental dimensions. Using the proposed framework, which is the ¯rst contribution of this paper, the suitability of the various candidate locations is assessed, after which a linear set-covering model is proposed to select a certain number of data centers. That model, which is the second contribution of this paper, guarantees both the suitability of the selected location and the coverage radius for high-quality services. Applying the proposed method on a large scale, in this case an entire country, is the third contribution of this study. The multi-criteria set-covering model was used to determine the best location(s) for data centers in Iran. At the moment, more than 30 servers are rented by Iranian companies in western countries, with the corresponding expenses for hosting and transmitting the data. Establishing data centers not only reduces the waiting time for Iranian users, but also it increases service reliability and security of information sharing on social networks. As such, setting up data centers is a logical step in expanding information technology in Iran.
The remainder of this paper is organized as follows. Section2contains a literature
review involving data center location, followed by a discussion of the framework and
the relevant criteria. In Sec. 3, the methodology being used is discussed, whereas
this study are addressed, and the most suitable data center locations in Iran are
selected. In Sec. 6, ¯nally, the conclusion of this study and suggestions for future
research are discussed.
2. Literature Review
Since few studies have focussed speci¯cally on selecting suitable locations for data centers, we examined papers about di®erent topics related to data centers to identify
relevant criteria. The results are presented in Table1. References and de¯nitions of
criteria were used to divide them into economic, social and environmental categories
(Table1). The papers related to the location of data centers are discussed below.
Chang et al.7solved a capacitated p-median problem to select the optimal location
of army data centers in the USA. The mathematical model they proposed was based on two types of decision drivers, namely, facility capacity and application perfor-mance. It minimizes both the total demand-weighted distance and the load on any
area processing centers. Abbasov et al.8 suggested a linear integer programming
model to determine the location of data centers. The proposed model determines the optimal locations based on possible natural, political and economic risk factors. Floods, earthquakes, pollution, magnetic radiation, government laws and regula-tions, inter-state relaregula-tions, political and social events, tax policies, economics and
dependence on the national economy are among the main criteria. Yang and Ye9
used the Delphi method to select the location for data centers of industrial agencies in China. Physical data security, su±cient resources and costs, environmental factors, government policies, manpower and level of economic are among the main criteria. The results indicate that natural geography is the most important criterion.
Goiri et al.10assessed the locations of data centers in seven regions in the USA,
using a nonlinear cost optimization model that minimizes the costs of network la-tency, consistency delay and availability constraints. The potential data center locations were identi¯ed based on (i) the distance to power plants, population centers and network backbones, (ii) the origin of electricity in the location, (iii) the price of water, land and electricity in the region and (iv) the temperature conditions in the region. They solved the suggested model using ¯ve di®erent approaches. Larumbe
and Sanso11proposed a convex integer programming to solve the cloud location and
routing problem for global content providers, including Google, Akamai, Amazon and Facebook. The objective function in the proposed model minimizes the average network delay, subject to system and budget constraints. Based on the results, 11 locations in Africa, Asia, Europe, Latin America and North America were selected
for data centers from among 24 potential locations. Covas et al.12applied
Elimina-tion and Choice Expressing Reality III (ELECTRE III) in a group decision-making environment for locating data centers in Portugal. They examined possible alter-natives using 35 criteria, divided into risk-related, social, economic and
environ-mental categories. Covas et al.13 used geographic information system (GIS) to
Table 1. Sustain able frame work for loca ting dat a center's loca tion. Cate gory Criteria Su b-criteria Refer ences Social Lif e qua lity (attr activeness to emplo yees) 13 Avai labi lity of pu blic tran sportation and access ibilities . Rail statio ns . Main roads . Acces s to airport 13 Lif e cost (mun icipalities tax discount s-IRS) 13 Lif e sec urity 9, 12, 14 . Crime against prope rty (crime against pe rsons) 9, 12 –14 . Polic e 1 2 . Fir eman 12 Leve l o f expan sion of society 9, 12 Avai labi lity of skill ed labor in the regi on 12, 13, 16 P olitical risks 8 C hange in the gen eral poli cies and laws of the country 8, 12 Ine± cient and incom plete laws 8 E xistence of restrictive laws 8 Go vernm ent failure to comp ly with obligatio ns 8 We akn ess of intern ational rela tions 8 Attra ctiveness to costumer (distan ce to data ce nter) 12 Dis tance to main road access 12 Dis tance to rail stations 12 Dis tance to airports 12 Info rmati on security 9 High -tech talent resources 9
Table 1. (Contin ued ) Cate gory Criteria Su b-criteria Refer ences Econom ical Inve stm ent costs 12, 20 E lectric grid conn ection cost 12, 14, 19 Lo cal inc entives 8 , 1 3 . Easie r licensing . Tax reduc tion . Acces s to subsid ies . Com petitive prices . Land for free 14 Net work comm unica tions conn ection cost s 7 , 1 2– 14, 18, 19 La nd acq uisition and constru ction cost 12, 14, 16, 19 Syn chroniza tion costs 12 Natu ral gas grid conn ection costs 12 Ope rational cost s 12, 16, 19 T axes 8, 9, 12, 14 E nergy costs 12 Hu man reso urces costs 9, 12, 13 C limate cond ition cost s . Hu midity . Tempe rature 11 –13 E conomic risks 8 E ®ective comm unica tion be tween investors and the mar ket 8 De pendence of the nation al economy 8 Avai labi lity of qua li¯ed supp ort vendor in the regi on 12 P hysical security 9 Mili tary regi onal sec urity 8, 9, 12 . Act ive de fensive . Nonac tive defensiv e . Front ier threats
Table 1. (Contin ued ) Cate gory Criteria Su b-criteria Refer ences Natu ral and ecological risks 8, 12 . Seismic activity 8, 12, 13 . Earth quake 8 . Oth er nat ural disa sters (stor ms, °oods and landslides) 8 . Elec tromagne tic radiation 8 . Pollu tion 8 . Inte rruptio n o f earth 8 Env ironme ntal Lo cal poll ution 8, 9, 12, 21 Noise 9 Re sidues T emperatu re 11 –13 Air poll ution 9, 12 E nergy savings 9, 12, 22 –24 Re newab le energ y sources (sola r and wind ) 1 2 Are as where waste heat data ce nter can be reu sed (o±ce bu ilding, swimm ing pool s, greenho uses) 12 P otentia l for fre e coo ling by wate r 14, 16 P otentia l for fre e coo ling by air 12, 16 Inte rferen ce with protec ted areas 13
Alternatives were evaluated based on eight criteria, divided into environmental,
economic and risk-related categories. Daim et al.14 proposed hierarchical decision
model (HDM) to select the best location of a data center in California. To assess the alternatives, 12 criteria were used, divided into ¯ve categories, including ¯nancial, environmental, social, political and geographic. Based on the results, telecom net-work availability, land cost, tax structure, incentives and subsidies and safety and security crime were identi¯ed as the main criteria in selecting the best location for the data center.
Klinkowski et al.15 focussed on the optimization algorithms that minimize the
required spectrum in the location selection problem of the data center. To this end, they ¯rst proposed a mathematical model to minimize the required spectrum based on demand constraints, after which they used several algorithms to solve the pro-posed model. The results indicated that the column generation technique presents
good results within a reasonable time frame. Ouni¯ et al.16 used a mixed integer
linear programming (MILP) model to select the optimal location of the data center from ¯ve areas in Quebec. The objective function in the proposed model minimizes di®erent costs, such as ¯xed and variable, server and switch and energy costs. To solve the MILP model, nine parameters were applied, including costs of maintenance, server, administration, energy, land and bandwidth. Based on the results, energy cost, accounting for 21% of the total costs, was determined to be the main factor in
determining the best location for the data center. Depoorter et al.17 conducted a
study to illustrate the e±ciency of the data center location in di®erent regions of Europe. To that end, they analyzed the data center's energy e±ciency with both direct air free cooling strategy and photovoltaic system integration to identify can-didate locations, using energy consumption, electricity generation, renewable energy supply, primary energy consumption, carbon emissions and energy cost as the main indicators. The results show that in comparison with photovoltaic system integra-tion, the direct free air-cooling strategy signi¯cantly reduces the energy consumption
of the data center. Mustafa et al.18used GIS and analytic hierarchy process (AHP) to
determine the most suitable locations for landslide monitoring in Cameron Highland, identifying tower distance, elevation, communication service coverage and slope as the main criteria. The results indicated that communication service coverage
has the greatest impact on the data center location selection problem. Liu et al.19
proposed a hybrid methodology, including the Lagrangian multiplier method and particle swarm optimization algorithm, to determine the optimal location of railway data centers in China. Minimizing the transmission distance between di®erent railway data centers is the main aim of their proposed method. They evaluated the locations based on factors like construction costs, availability of spe-cialists in the region, existing communication networks and electricity e±ciency (maintenance).
As the literature review shows, both MCDM methods and mathematical pro-gramming are popular when it comes to determining the optimal location of data centers. The review reveals that (i) although many criteria have been identi¯ed in
literature as a®ecting the location of data centers, there is no comprehensive framework bringing those criteria altogether, (ii) the geographical distribution of the selected locations has been ignored in most MCDM studies and (iii) the mathe-matical models suggested in literature may not cover all criteria (especially the qualitative ones) that a®ect the location of data center. This may have to do with the fact that, by increasing the number of constraints and variables, not only does
the complexity of creating a mathematical model increase,25 but the feasibility
of the mathematical model in question is also reduced dramatically. Therefore, it turns out that the application of MCDM or a mathematical model, as a single solution to the problem of ¯nding the best location for data centers, may not be su±ciently e®ective.
To remedy that state of a®airs, a multi-criteria set-covering method is proposed in this paper, consisting of two parts, an MCDM method, namely BWM, to calculate the utility of potential locations and a set-covering model to determine the optimal location of data centers. We used the performance of each alternative in the criteria
presented in Table1to calculate the utility of each alternative presented. The results
of the multi-criteria part are then used as parameters of the objective function in the model suggested in the second part of the proposed methodology. Using the set-covering model not only addresses the problem of the geographical distribution of
selected locations but also means that a wide range of criteria, presented in Table1,
is considered in the selection process.
3. Research Methodology
As shown in Fig.1, the methodology used in this paper consists of two parts and ¯ve
steps. In the ¯rst part called MCDM, the utility of alternatives (in this case the provinces of Iran) is calculated through three steps. To this end, in the ¯rst step a
sustainable framework of criteria is created through literature review (Table1) and
then using the best–worst method (BWM) and additive value function the weight of
the criteria and the utility of each alternative are computed in Steps 2 and 3, respectively.
In the second part of the methodology applied in this research, ¯rst using the result of additive value function, a linear set-covering model is proposed in Step 4 and then through the brute-force and genetic algorithm the developed model is evaluated in Step 5.
3.1. BWM
As presented in Fig.1, to calculate the weight of criteria, presented in Table1, BWM
is employed in this research. BWM, which is a pairwise comparison-based method,
has many advantages compared to weighting methods.26–28The number of pairwise
The high consistency rate provided by BWM makes its results more reliable than
similar methods.26,29,30 BWM has been successfully used in several real-world
pro-blems, including location selection propro-blems,31–33sustainability,34–36technology
se-lection,37,38 emergency decision-making,39 reliability engineering,40 customer
requirements.41and supply chain management.42–44The process of weighting using
BWM is divided into ¯ve steps, as follows.26,27
(1) Determine a set of decision criteriafc1; c2; . . . ; cng. This step is done by
decision-makers and/or experts.
(2) Determine the best (B) and the worst (W ) criteria by the decision-makers and/ or experts.
(3) Determine the preference of the best criterion (B) over all the other criteria by the decision-makers and/or experts using a number from 1 to 9 (where 1 is equally important and 9 is extremely more important). The result of
\Best-to-Others" comparisons is the vector AB ¼ ðaB1; aB2; . . . ; aBj; . . . ; aBnÞ, where aBj
indicates the preference of criterion B over criterion j
(4) Determine the preference of all the criteria over the worst by the decision-makers and/or experts using a number from 1 to 9. The result of \Others-to-Worst"
comparisons is the vector Aw¼ ða1W; a2W; . . . ; ajW; . . . ; anWÞ, where ajW
denotes the preference of criterion j over criterion W:
(5) Calculate the optimal weightsðw1; w2; . . . ; wnÞ by an optimization model.
The optimal weights are calculated by minimizing the maximum offjwB aBjwjj;
jwj ajWwWjg for all j, which is translated into the following optimization model:
min max j fjwB aBjwjj; jwj ajWwWjg subject to: Xn j¼1 wj ¼ 1 wj 0; for all j: ð1Þ
Model (1) is transferred into:
min subject to: jwB aBjwjj , for all j jwj ajWwWj , for all j Xn j¼1 wj ¼ 1 wj 0; for all j: ð2Þ
Model (2) is used to determine the local weight of the criteriaðw1; w2; . . . ; wnÞ and
the consistency indicator ðÞ. When there is more than one level in the
decision-making hierarchy, the local weights are ¯rst calculated using Model (2), after which the global weights of criteria are computed by multiplying the local weights at the
di®erent levels. After determining the global weight of the criteria, by using Eq. (3),
the additive value function presented by Keeney and Rai®a,45 the utility of
alter-natives with regard to di®erent criteria, is calculated46,47:
Vi¼
X j
wjuij for all i; ð3Þ
where wj and uij are the global weight of criterion j and the normalized value of
alternative i in criterion j, respectively using Eqs. (4) and (5), the amount of uij
for monotonic increasing (e.g. quality of life) and monotonic decreasing (e.g. cost) criteria can be calculated.
uij¼ xij
maxiðxijÞ
for all i and j; ð4Þ
uij ¼ min
iðxijÞ
xij
3.2. Linear set-covering model
In the second part of the methodology, the most suitable locations for data centers are identi¯ed using the linear set-covering model. The utility of the potential can-didates calculated in the ¯rst part, the number of data centers and the coverage radius of each data center are included as parameters in the proposed model. We used the following notations for developing the mathematical model.
Notations
P¼ fp1; p2; . . . ; png A set of n provinces
Dis : P P ! R>0 A distance function between each pair of provinces
Per : P ! ½0; 1 Normalized performance function which assigns a
proper performance showing how much a province deserves to be a center without considering utility or distance of other provinces
T A distance threshold (i.e., the cover radius for
each data center)
k Maximum number of provinces which can be selected
as a data center (this can also be replaced by budget constraints)
xi A binary decision variable: if a province pi is selected as a
data center place xi¼ 1, otherwise xi¼ 0
X ¼ fx1; x2; . . . ; xng Solution set
D½i; j; A Boolean n n matrix shows the distance between
each pair of provinces is less than T or
not. D½i; j ¼ 1 shows Dðpi; pjÞ T, otherwise D½i; j ¼ 0:
Vj The utility of province pi resulted by MCDM (Eq. (3))
Conditional objective function
The proposed linear set-covering model, including conditional performance function and constraints, is described below.
Conditional performance functionPerðpijXÞ
According to the objective function (Eq. (6)), the provinces selected for a data center
not only cover the service applicants in a speci¯c radius but also have a high per-formance in the various criteria, divided into economic, social and environmental categories. Since choosing a province for a data center a®ects the utility of the other provinces, we need an approach to handle such interactions. In this regard, we introduce a simple and e±cient method, called conditional performance function. The idea is to gradually reduce the utility of a province with regard to the distance
and utility value of the surrounding data centers. To this end, we de¯neði; jÞ as the
distance interaction function, where ði; jÞ ¼ 0 shows the distance of provinces pi
and pjis far from the distance threshold T , which means that there is no interaction
such that by reducing the distance between provinces piand pj, the value ofði; jÞ is
gradually increased. ðiÞ is an aggregative utility function for the province pi. It
denotes the aggregative impact of surrounding provinces of piand reduces the utility
of a province piup to half of its independent utility. If, in a solution vector X, there is
no data center near pi,ði; jÞ ¼ 0 and consequently it results in ðiÞ ¼ 0, that is, the
conditional performance function Perðpij XÞ ¼ Vj. Otherwise, Perðpij XÞ ¼Vj
1þðiÞ
is a reduced or conditional utility value of pi. The method is formulated as follows.
Maximize PerðXÞ ¼X
n
i¼1
xiPerðpij XÞ: ð6Þ
In the conditional performance functionPerðpijXÞ :
ði; jÞ ¼ 0; if Dis½i; j > T; T Dis½i; j T ; otherwise; 8 < : ðiÞ ¼ Pnj¼1 j6¼i xjði; jÞVj Pn j¼i xjD½i; jVj ; Perðpij XÞ ¼ Vj 1þ ðiÞ: ð7Þ Constraints
As indicated in constraint (8), the number of selected provinces should be equal to or less than k:
Xn
i¼1
xi k: ð8Þ
According to constraint (9), for each province pi, there is at least one province pj
such that xj ¼ 1 and Disðpi; pjÞ T:
Xn
j¼1
xjD½i; j > 0; 8i ¼ 1; 2; . . . ; n ð9Þ
4. Data Collection
In this section, we discuss the procedure we used to collect data to evaluate the various provinces of Iran as possible data center locations. The data collection process for this study consisted of two steps, with data being collected to weight the criteria and calculate alternatives scores. The steps are discussed in greater detail below.
4.1. Data for the criteria
In the ¯rst data collection step, ¯rst the criteria included in Table1were screened.
discriminant power of decision maker(s)/experts48 and the inconsistency in the
pairwise comparisons49are decreased. To screen the criteria, 11 experts were invited
to ¯ll in an online questionnaire based on a ¯ve-point Likert scale. After aggregating the experts' opinion, the number of 3 (out of 5) was selected as a threshold, which leads to appropriate balance among the three dimensions of hierarchical tree. The
results are presented in Fig.2.
Then, BWM was used to design an online questionnaire for weighting the criteria listed above, in which 35 experts were asked to provide their opinions. All respondents in both stages have academic knowledge about and work experience with data centres. They were identi¯ed by their online pro¯le. Since Iran is a large country with di®erent economic, social and environmental conditions, we have tried to select the experts from di®erent parts of Iran. So that, of 35 experts invited in this research, 11 (31%) worked in east provinces of Iran and the remaining experts were employed as specialist members of server room in center (23%), west (17%), north (15%) and south (14%) of
Iran. Table2shows the speci¯c backgrounds of the respondents used in both steps. For
aggregating the experts' opinion in the weighting step, geometric mean is applied. 4.2. Data for the alternatives
In the second step of data collection, based on the type of criteria (i.e., qualitative or quantitative), data were collected to evaluate the utility of provinces of Iran as
alternatives. From among the criteria presented in Fig.2, local incentives have been considered as qualitative criteria. The score of provinces of Iran, in this criterion, was determined based on the experts' opinion through an online questionnaire. To this end, we used a 10-point Likert scale to determine the score of provinces of Iran for local incentives criterion. For collecting data in other criteria, both web sites and internal database of Iran Meteorological Organization, the Law Enforcement Force of Iran, the Ministry of Science Research and Technology, the Ministry of Culture and Islamic Guidance, the Ministry of Health and Medical Education, the Statistical Centre of Iran, the Ministry of Housing and Urban Development and the Institute for Research and Planning in Higher Education were used in this research.
5. Results and Discussion
In this section, ¯rst the local weight of the criteria shown in Fig. 2 is analyzed.
Secondly, the global weights and data collected from di®erent sources are used to determine the suitability of the di®erent provinces of Iran for the location of a data center. Finally, the performance of the linear set-covering model is evaluated based on a number of examples.
5.1. Local weight of criteria
A sustainable framework of criteria is applied that includes social, economic and environmental dimensions to determine the optimal location for a data center in
Iran. The mean values of criteria for the three levels are presented in Tables3–5.
According to the ¯ndings, which are based on expert opinions, the economic criteria
are more important than either social or environmental criteria (Table3). A lack of
transparency when it comes to supporting investors in a developing country like Iran, the general instability in the Middle East, which limits investments into sensitive facilities like data centers, and the unstable economic conditions of Iran as a result of
the sanctions in particular50,51could explain the distribution of the weights.
Table 2. Speci¯cations of experts.
Respondents For screening criteria Average years of work experience For weighting criteria Average years of work experience
Information and Communication Center of Ferdowsi University
11 8.5
Research Center of Information Technology of Iran University of Science and Technology
6 10.6
Information and Communication Center of Vali-E-Asr University of Rafsanjan
4 8.4
Tehran Municipality ICT Organization 2 5 Specialists in di®erent server rooms 11 3.81 12 5.5
Economic dimension
Of the indicators in the economic category, investment costs were weighted as the
most important criterion by the experts (Table4). The cost of equipment, problems
with regard to importing the equipment to Iran because of the sanctions,52,53the high
exchange rate in Iran54 and the lack of transparency facing investors in Iran55are
some of the reasons for that. Operational costs and economic risks were ranked
second and third, with not much di®erence between them (Table4).
Since the problems mentioned in the investment section can be remedied by
providing incentives like free land, taxes and subsidies,55the experts considered local
incentivesas the most important sub-criterion in the category of investment costs
(Table 5). Land acquisition and construction costs, network connectivity costs,
electric grid connection cost and synchronization costs are other important factors in
this category (Table 5). The earthquake has been selected as the most important
indicator of the economic risk category by experts (Table 5). The geographical
location of Iran, in an earthquake zone, justi¯es that selection. Electromagnetic radiation and other natural disasters were also considered to be important
sub-criteria in this category (Table5).
Social dimension
Of the three criteria included in the social category, information security is selected
as the main factor (Table4). The role of Iran in the political, economic and military
constellation in the Middle East56combined with high levels of insecurity in
neigh-boring countries57are reasons to select information security as an important factor in
this regard. Life quality and political risks are two other important criteria in this
Table 3. Weigh of the main criteria.
Criteria Weight Rank
Economic 0.416 1 Environmental 0.324 2
Social 0.260 3
Table 4. Weigh of sub-criteria in level 2.
Category Sub-criteria Weight Rank
Economic Investment costs 0.380 1 Operational costs 0.316 2 Economic risks 0.304 3
Social Life quality 0.317 2
Political risks 0.259 3 Information security 0.424 1 Environmental Local pollution 0.374 2
Energy saving 0.388 1
category (Table4). The results of the analysis in the weight of criteria categorized into political risks indicate that changes in the general policies and laws of the country, with very little di®erence between them, are more important than the ex-istence of restrictive laws and government failure to comply with obligations
(Table5). The unstable economic condition in Iran can be mentioned to justify the
high weight of the ¯rst two criteria. In an economic crisis, di®erent restrictive laws and policies could be implemented and cause social upheaval, like a reduction in
the workforce,58which would a®ect society directly. Based on the expert opinions, of
the criteria included in life quality, life security is more important with regard to the location of a data center than the availability of public transport and accessibilities
(Table5). The size of the country, which can lead to a reduction in security in some
areas, combined with the high levels of insecurity in neighboring countries57 are
among the reasons why the experts considered life security to be that important. Environmental dimension
Of the criteria included in the ¯rst level of the environmental category, energy saving, local pollution and interference with the protected area were identi¯ed by the
experts as being the most important criteria (Table4). This may have been a®ected
by air pollution levels in Iran. Due to the heat generated by data centers, there may be some areas of the country that would be unable to meet the energy demand
involved,59 while the use of fossil fuels, the main source of air pollution in Iran,
is reduced. According to the experts, the potential for free cooling is the most
important sub-criterion of energy saving (Table 5). The high cost of cooling the
equipment60and di®erences in climate condition in various parts of Iran are among
the reasons for assigning the high weight to the potential for free cooling. Renewable
Table 5. Weigh of sub-criteria in level 3.
Category Sub-criteria Weight Rank
Investment costs Electric grid connection cost 0.198 4
Local incentives 0.247 1
Network communications connection costs 0.199 3 Land acquisition and construction cost 0.205 2 Synchronization costs 0.152 5
Economic risks Earthquake 0.400 1
Other natural disasters (storms, °oods and landslides) 0.265 3 Electromagnetic radiation 0.334 2 Life quality Availability of public transportation and accessibilities 0.281 2
Life security 0.719 1
Political risks Change in the general policies and laws of the country 0.364 1 The existence of restrictive laws 0.363 2 Government failure to comply with obligations 0.273 3 Energy saving Renewable energy source (solar, wind) 0.386 2 Areas where waste heat data center can be reused 0.224 3 Potential for free cooling 0.391 1
energy sources (solar, wind) and areas where waste heat data center can be reused
were considered to be the other two important factors in this category (Table5).
5.2. Alternatives' utility
To determine the utility of provinces of Iran for establishing a data center, ¯rst the
global weight of criteria, wj, is calculated (Table6). Based on wj in Table 6,
oper-ational costs, local pollution and information security, respectively, are the ¯rst three criteria that play a signi¯cant role in deciding the suitability of alternatives.
By using wj (mean values) and uij as shown in Table A.1 in Appendix A, the
suitability of candidate locations is computed by Eq. (3) for the various provinces of
Iran (Table7). The results indicate that, of the 31 provinces in Iran, Qom, Semnan
and Zanjan o®er the best alternatives. However, as shown in Fig. 3, the three
alternatives are located near each other, which may a®ect the service for clients further a¯eld (i.e., customers located in the southern and south eastern provinces of Iran). 5.3. Results of the multi-criteria set-covering model
As discussed earlier, to ensure an appropriate geographical distribution in the se-lected location of data centers, something that is not addressed by MCDM methods, we use a multi-criteria set-covering method, using the results of the MCDM part as parameters in the objective function of the linear set-covering model. By using the proposed model, the selected centers guarantee both a good performance on the criteria in the economic, social and environmental categories and an appropriate
Table 6. Global weight of sub-criteria.
Criteria Global weightðwjÞ Rank
Operational costs 0.132 1
Local pollution 0.121 2
Information security 0.110 3
Interference with protected areas 0.077 4
Life security 0.059 5
Earthquake 0.051 6
Potential for free cooling 0.049 7 Renewable energy source (solar and wind) 0.048 8
Electromagnetic radiation 0.042 9
Local incentives 0.039 10
Other natural disasters (storms, °oods and landslides) 0.034 11 Land acquisition and construction 0.033 12 Network communications connection costs 0.032 13 Electric grid connection cost 0.031 14 Areas where waste heat data center can be reused 0.028 15 Change in the general policies and laws of the country 0.025 16 The existence of restrictive laws 0.0241 17
Synchronization costs 0.024 18
Availability of public transportation and accessibilities 0.023 19 Government failure to comply with obligation 0.018 20
geographical distribution, which guarantees high-quality services for potential cli-ents. To demonstrate the performance of the proposed model, we analyze the
problem under di®erent conditions (Table8). To analyze the problem, two methods,
Brute-Force and a genetic algorithm, were applied. Although the Brute-Force method provides optimal solutions, it is not a practical method for the large instances that exist in a country like Iran. In this regard, we used Brute-Force method to solve
problems with small number of centers (i.e., k 8), solving the rest using a genetic
algorithm (Table 8). Note that the parameter D½i; j, which shows the distance
between alternatives, is presented in Table A.2 in Appendix A.
As Table8indicates, there is a connection between the number of the data center
(s), the coverage radius of data centers in a speci¯c area and the geographical dis-tribution of selected locations. In other words, when there is a limitation on the number of data centers, for instance, due to a limited budget, the suggested model tries to select locations that have a high geographical dispersion for a small coverage
Table 7. The utility of Iran's provinces for data center location.
Provinces of Iran Utility of candidate places Rank
Qom 0.4645 1 Semnan 0.4603 2 Zanjan 0.4542 3 Markazi 0.4444 4 South Khorasan 0.4416 5 Bushehr 0.4320 6 Qazvin 0.4272 7 Ardabil 0.4197 8 North Khorasan 0.4081 9 Hormozgan 0.4050 10 Yazd 0.3991 11 Alborz 0.3985 12 Tehran 0.3984 13
Kohgeluyeh and Boyer-Ahmad 0.3954 14
Ilam 0.3945 15
Razavi Khorasan 0.3821 16
Isfahan 0.3617 17
Kordestan 0.3434 18
Chaharmahal and Bakhtiari 0.3419 19
Kerman 0.3416 20
Hamadan 0.3375 21
East Azarbaijan 0.3341 22
Mazandaran 0.3304 23
Sistan and Baluchestan 0.3289 24
Gilan 0.3232 25 Kermanshah 0.3150 26 Golestan 0.3017 27 Fars 0.2973 28 Lorestan 0.2741 29 West Azarbaijan 0.2735 30 Khuzestan 0.2714 31
radius. Otherwise, the suitability of the selected places is the ¯rst priority through the
proposed model for a large coverage radius. Figure 4 clari¯es the case mentioned for
three data centers. As indicated in Fig.4, for T ¼ 800 km and 1000 km, only Zanjan,
which is a province in the north-west of Iran, is selected as a highly suitable location.
However, when the coverage radius of each data center increases (i.e., T ¼ 1200 km and
1500 km) and taking into account that not all potential clients may receive high-quality
services, the most suitable locations (i.e., Qom, Semnan and Zanjn) are selected (Fig.4).
In contrast, when there is no limitation on the number of the selected locations, the results provided by the set-covering model yield about the same results as the MCDM, even for small coverage radius, which means that the role of the utility in the selection becomes far more important than the distribution of data centers. In other words, by increasing the number of data centers, all potential clients can be serviced by at least one data center within a reasonable distance, so the objective function in the proposed model attempts to select the most suitable locations even for small
coverage radius. For instance, as shown in Table 8, when T¼ 500 km, when we
increase the number of data centers, the most suitable locations (i.e., Qom, Semnan and Zanjan) are selected as alternatives, which means that based on the opinion of decision-maker(s)/experts who want to know the suitable locations of data center, if the suitability of the selected locations is more important than the coverage radius of data centers, or vice versa, the linear set-covering model performs well.
Tabl e 8 . A n alyzing the proble m u n der di®ere nt sett ings. Selec ted places for (T is dist ance in km) Solv ed by T ¼ 500 T ¼ 800 T ¼ 1000 T ¼ 1200 T ¼ 1500 Brut e-For ce k ¼ 1 N o feasib le solution No feasible solu tion No feasib le solution No fea sible soluti on Qom Objec tive func tion: 0.464 5 k ¼ 3 N o feasib le solution Zanjan Zanj an Qom Qom South Khorasa n South Khorasa n Semnan Semn an Kohge luyeh and Boye r-Ahm ad Bush ehr South Kho rasan Zanjan Objective function : 1.291 2 Objec tive fu nction: 1.327 8 Objective function : 1.366 4 Objec tive func tion: 1.379 0 k ¼ 5 N o feasib le solution Qom Qom Qom Qom Semnan Semn an Semnan Semn an Zanjan Zanj an Zanjan Zanjan South Khorasa n Bush ehr Markazi Mark azi Yazd South Khorasa n South Kho rasan South Khorasa n Objective function : 2.219 7 Objec tive fu nction: 2252 6 Objective function : 2.265 0 Objec tive func tion: 2.265 0 k ¼ 8 Qom Qom Qom Qom Qom Se mnan Semnan Semn an Semnan Semn an Za njan Zanjan Zanj an Zanjan Zanjan Sou th Kho rasan Mark azi Mark azi Markazi Mark azi B ushehr South Khorasa n South Khorasa n South Kho rasan South Khorasa n Nort h Khorasa n Bushe hr Bush ehr Bushe hr Horm ozgan Qazvin Qaz vin Qazvin Qaz vin Kord estan Hormoz gan Ard abil Ardab il Ardab il
Table 8. (Contin ued ) Selec ted places for (T is dist ance in km) Solv ed by T ¼ 500 T ¼ 800 T ¼ 1000 T ¼ 1200 T ¼ 1500 Obj ective function : 3.4091 Objective function : 3.529 2 Objec tive fu nction: 3.543 9 Objective function : 3.543 9 Objec tive func tion: 3.543 9 Gene tic algor ithm k ¼ 10 Qom Qom Qom Qom Qom Se mnan Semnan Semn an Semnan Semn an Za njan Zanjan Zanj an Zanjan Zanjan Mark azi Mark azi Mark azi Markazi Mark azi Sou th Kho rasan South Khorasa n South Khorasa n South Kho rasan South Khorasa n B ushehr Bushe hr Bush ehr Bushe hr Bushe hr Nort h Khorasa n Qazvin Qaz vin Qazvin Qaz vin Horm ozgan Ardab il Ard abil Ardab il Ardab il Yaz d Nort h Kho rasan Nort h Khorasa n North Khorasa n Nort h Kho rasan Obj ective function : 4.2526 Objective function : 4.351 1 Objec tive fu nction: 4.357 0 Objective function : 4.357 0 Objec tive func tion: 4.357 0 k ¼ 12 Qom Qom Qom Qom Qom Se mnan Semnan Semn an Semnan Semn an Za njan Zanjan Zanj an Zanjan Zanjan Mark azi Mark azi Mark azi Markazi Mark azi Sou th Kho rasan South Khorasa n South Khorasa n South Kho rasan South Khorasa n B ushehr Bushe hr Bush ehr Bushe hr Bushe hr Qaz vin Qazvin Qaz vin Qazvin Qaz vin Ard abil Ardab il Ard abil Ardab il Ardab il Nort h Khorasa n Nort h Kho rasan Nort h Khorasa n North Khorasa n Nort h Kho rasan Horm ozgan Hormoz gan Horm ozgan Hormoz gan Hormoz gan Ila m Alborz Yaz d Yazd Yazd Kord estan Tehran Alb orz Alborz Alb orz Obj ective function : 5.0949 Objective function : 5.153 9 Objec tive fu nction: 5.154 6 Objective function : 5.154 6 Objec tive func tion: 5.154 6
6. Conclusion and Future Research
In this paper, a multi-criteria set-covering method was used to determine the location of data centers. In the proposed method, ¯rst the utility of candidate places was calculated, using the BWM, after which a linear set-covering model was used to select the optimal number of locations for data centers that cover a speci¯c radius. To evaluate the suitability of candidate locations, sustainable framework of criteria was created on the basis of a comprehensive literature review. The proposed meth-odology was applied to Iran. To weighting the criteria included in the proposed framework, a group of specialists from Iran was asked to ¯ll in an online question-naire designed based on the BWM, the results indicating that operational costs, local pollution and information security are the most important criteria, respectively. The weight of the criteria was then used to determine the suitability of the various provinces of Iran.
In addition to the criteria included in the proposed framework, the layout of data centers plays an important role in the quality of the services provided by the various data centers, because the speed of accessing to information in cyberspace is a®ected by the radius covered by the data centers. In this regard, by applying the suitability of the alternative locations and the coverage radius of each data center as para-meters, a linear set-covering model was used to select a certain number of data centers in the second part of the proposed methodology, using a brute-force algo-rithm and a genetic algoalgo-rithm to solve the proposed model. The performance of the proposed model was tested under di®erent settings.
The sustainable framework proposed in this paper has a number of advantages. Using the list of the criteria presented in the framework, public policymakers can produce better laws to increase the level of satisfaction regarding Internet services, while decision-makers can improve the quality of the selected locations.
Since the evaluation of candidate places should satisfy two major requirements: sustainability perspective and geographical distribution, which could apply to many other similar location selection problems (e.g., telecommunication or biofuel distri-bution center selection), the use of the proposed multi-criteria set-covering meth-odology in those similar problems could be considered as one possible avenue for future research.
We also suggest including other features of data centers, such as workload/tra±c exchanged within a data center network, as well as using other optimization algo-rithms such as simulated annealing, particle swarm optimization and tabu search to solve the suggested model.
Appendix A Table A.1. Normalized value of alternatives in di®erent criteria. Criteria Alternatives Operational cost Local pollution Information security Interference with protected areas Life security Earthquake Potential for free cooling Renewable energy source (solar, wind) Electromagnetic radiation Local incentives (obtaining of construction license) Other natural disaster (storms, °oods and landslides) East Azarbaijan (A 1 ) 0.3258 0.1484 0.2448 0.4366 0.3631 0.0950 0.7686 0.1369 0.8184 1.0000 0.0714 West Azarbaijan (A 2 ) 0.2537 0.1777 0.2615 0.1047 0.3462 0.1275 0.8304 0.0995 0.7585 0.4941 0.0435 Ardabil (A 3 ) 0.4559 0.4567 0.7333 0.1163 0.1247 0.5135 1.0000 0.3831 0.9426 0.4941 0.1176 Isfahan (A 4 ) 0.5841 0.1133 0.2285 0.2707 0.3972 0.2159 0.5706 0.9119 0.7226 0.3216 0.0606 Alborz (A 5 ) 0.8240 0.2139 0.5220 0.0654 0.1753 0.6129 0.6039 0.1358 0.7870 0.5059 0.2000 Ilam (A 6 ) 0.3077 1.0000 0.7700 0.1321 0.0521 0.0603 0.5439 0.0630 0.7631 0.4941 0.1538 Bushehr (A 7 ) 0.2173 0.4987 0.8415 0.1365 0.0952 0.0417 0.3750 0.1170 0.7072 1.0000 0.0417 Tehran (A 8 ) 0.7156 0.0437 0.0917 0.3377 1.0000 0.3455 0.5376 0.0177 0.7870 0.3216 0.1111 Chaharmahal and Bakhtiari (A 9 ) 0.3133 0.6121 0.6364 0.1379 0.1444 0.1387 0.7623 0.0409 0.7623 0.3216 0.1111 South Khorasan (A 10 ) 0.3773 0.7545 1.0000 0.0849 0.0910 0.3167 0.5536 0.5897 0.7267 0.5059 0.0303 Razavi Khorasan (A 11 ) 0.4546 0.0902 0.1548 0.7089 0.5719 0.2676 0.6596 0.6657 0.7834 0.5059 0.0426 North Khorasan (A 12 ) 0.4517 0.6722 0.6968 0.2439 0.1310 0.4872 0.7045 0.1273 0.8380 0.3216 0.1667 Khuzestan (A 13 ) 0.0632 0.1232 0.1812 0.3111 0.2693 0.0285 0.3563 0.2054 0.7372 1.0000 0.0171 Zanjan (A 14 ) 0.9185 0.5486 0.2388 0.1935 0.3707 1.0000 0.8087 0.1559 0.8121 0.3216 0.2857 Semnan (A 15 ) 0.3648 0.8260 0.7333 1.0000 0.1251 0.0950 0.5110 0.3469 0.8053 0.3216 0.0833 Sistan and Baluchestan (A 16 ) 0.1776 0.2091 0.5404 0.5339 0.1698 0.0826 0.5000 0.7315 0.6982 0.4941 0.0364 Fars (A 17 ) 0.1151 0.1196 0.1990 0.6889 0.4570 0.0138 0.5167 0.2442 0.7032 0.5059 0.0339 Qazvin (A 18 ) 0.9639 0.4555 0.4968 0.0228 0.1571 0.4222 0.6549 0.3141 0.8878 0.3216 0.2500 Qom (A 19 ) 0.7361 0.4489 0.7778 0.0279 0.1182 0.8636 0.5082 0.3425 0.7641 0.5059 1.0000 Kordestan (A 20 ) 0.4176 0.3619 0.1265 0.1587 0.5778 0.5135 0.6503 0.2593 0.7769 0.4941 0.3333
Table A.1. (Continued ) Criteria Alternatives Operational cost Local pollution Information security Interference with protected areas Life security Earthquake Potential for free cooling Renewable energy source (solar, wind) Electromagnetic radiation Local incentives (obtaining of construction license) Other natural disaster (storms, °oods and landslides) Kerman (A 21 ) 0.1212 0.1833 0.2862 0.7424 0.3097 0.0205 0.5471 1.0000 0.7066 0.5059 0.0278 Kermanshah (A 22 ) 0.3623 0.2971 0.4375 0.1090 0.1190 0.1508 0.6242 0.3009 0.7585 0.5059 0.1250 Kohgeluyeh and Boyer-Ahmad (A 23 ) 0.2949 0.8136 0.7064 0.1647 0.0966 0.0782 0.6370 0.1003 0.7596 0.3216 0.1176 Golestan (A 24 ) 0.2609 0.3104 0.4889 0.0445 0.1734 0.1367 0.5314 0.0663 0.7641 1.0000 0.0769 Gilan (A 25 ) 0.3313 0.2292 0.3298 0.1124 0.2627 0.1557 0.5776 0.1163 1.0000 1.0000 0.0385 Lorestan (A 26 ) 0.2236 0.3295 0.3675 0.1523 0.2072 0.0941 0.5376 0.1964 0.7216 0.3216 0.0667 Mazandaran (A 27 ) 0.3695 0.1767 0.4219 0.3660 0.2111 0.1532 0.5345 0.0507 0.9665 0.5059 0.0769 Markazi (A 28 ) 1.0000 0.4059 0.4952 0.0955 0.1805 0.5588 0.6788 0.2880 0.7657 0.5059 0.1000 Hormozgan (A 29 ) 0.2841 0.3266 0.3784 0.6375 0.2400 0.0204 0.3419 0.4490 0.7142 1.0000 0.0952 Hamadan (A 30 ) 0.4484 0.3338 0.5049 0.0548 0.1797 0.5135 0.7750 0.0826 0.7640 0.3216 0.0741 Yazd (A 31 ) 0.3886 0.5096 0.7624 0.4078 0.1179 0.0927 0.4745 0.4511 0.7132 0.5059 0.0769
Table A.1. (Continued ) Criteria Alternatives Land acquisition and construction Network communications connection costs Electric grid connection cost Areas where waste heat data center can be reused Change in the general policies and laws of the country The existence of restrictive laws Synchronization costs Availability of public transportation and accessibilities Government failure to comply with obligation East Azarbaijan (A 1 ) 0.1590 0.7137 0.1563 0.2947 0.1218 0.1218 0.1218 0.6076 0.1218 West Azarbaijan (A 2 ) 0.2459 0.6011 0.1563 0.2461 0.0662 0.0662 0.0662 0.5049 0.0662 Ardabil (A 3 ) 0.2358 0.6364 0.1563 0.0958 0.0825 0.0825 0.0825 0.3325 0.0825 Isfahan (A 4 ) 0.1149 0.7951 0.1049 0.3860 0.1776 0.1776 0.1776 0.6248 0.1776 Alborz (A 5 ) 0.1590 0.7395 0.5313 0.2044 0.1218 0.1218 0.1218 0.0815 0.1218 Ilam (A 6 ) 0.2400 0.6906 0.0759 0.0437 0.0868 0.0868 0.0868 0.1821 0.0868 Bushehr (A 7 ) 0.3173 0.7612 0.1272 0.0877 1.0000 1.0000 1.0000 0.2513 1.0000 Tehran (A 8 ) 0.0189 1.0000 1.0000 1.0000 0.1536 0.1536 0.1536 0.1626 0.1536 Chaharmahal and Bakhtiari (A 9 ) 0.1781 0.6065 0.1496 0.0714 0.0707 0.0707 0.0707 0.1855 0.0707 South Khorasan (A 10 ) 0.2977 0.6214 0.0246 0.0580 0.0832 0.0832 0.0832 0.6981 0.0832 Razavi Khorasan (A 11 ) 0.2682 0.6445 0.1138 0.4850 0.0858 0.0858 0.0858 0.8455 0.0858 North Khorasan (A 12 ) 0.2863 0.5360 0.0804 0.0651 0.0982 0.0982 0.0982 0.2305 0.0982 Khuzestan (A 13 ) 0.2350 0.6662 0.1272 0.3550 0.2888 0.2888 0.2888 0.8631 0.2888 Zanjan (A 14 ) 0.1966 0.6296 0.1362 0.0797 0.0919 0.0919 0.0919 0.3293 0.0919 Semnan (A 15 ) 0.1707 0.8955 0.0246 0.0529 0.1155 0.1155 0.1155 0.1917 0.1155 Sistan and Baluchestan (A 16 ) 0.2249 0.3731 0.0402 0.2092 0.0797 0.0797 0.0797 0.9658 0.0797
Table A.1. (Continued )( Continued ) Criteria Alternatives Land acquisition and construction Network communications connection costs Electric grid connection cost Areas where waste heat data center can be reused Change in the general policies and laws of the country The existence of restrictive laws Synchronization costs Availability of public transportation and accessibilities Government failure to comply with obligation Fars (A 17 ) 0.1625 0.7463 0.1049 0.3656 0.1297 0.1297 0.1297 1.0000 0.1297 Qazvin (A 18 ) 0.1451 0.6784 0.1674 0.0960 0.1226 0.1226 0.1226 0.2872 0.1226 Qom (A 19 ) 0.1347 0.7992 0.1451 0.0974 0.0866 0.0866 0.0866 0.0872 0.0866 Kordestan (A 20 ) 0.2405 0.5984 0.1183 0.1208 0.0846 0.0846 0.0846 0.3780 0.0846 Kerman (A 21 ) 0.2838 0.5807 0.0580 0.2385 0.1787 0.1787 0.1787 0.7655 0.1787 Kermanshah (A 22 ) 0.3001 0.5997 0.1585 0.1472 0.1169 0.1169 0.1169 0.4510 0.1169 Kohgeluyeh and Boyer-Ahmad (A 23 ) 1.0000 0.6092 0.1161 0.0537 0.1056 0.1056 0.1056 0.2742 0.1056 Golestan (A 24 ) 0.3488 0.6038 0.1563 0.1409 0.0712 0.0712 0.0712 0.2716 0.0712 Gilan (A 25 ) 0.2620 0.6947 0.4174 0.1907 0.0659 0.0659 0.0659 0.5218 0.0659 Lorestan (A 26 ) 0.3949 0.5712 0.1161 0.1327 0.0652 0.0652 0.0652 0.4402 0.0652 Mazandaran (A 27 ) 0.2606 0.8670 0.3348 0.2475 0.1087 0.1087 0.1087 0.4457 0.1087 Markazi (A 28 ) 0.1440 0.7313 0.1473 0.1077 0.1794 0.1794 0.1794 0.3424 0.1794 Hormozgan (A 29 ) 0.2748 0.6811 0.0759 0.1339 0.6887 0.6887 0.6887 0.5166 0.6887 Hamadan (A 30 ) 0.2503 0.6201 0.2031 0.1310 0.0762 0.0762 0.0762 0.3156 0.0762 Yazd (A 31 ) 0.2956 0.8657 0.0513 0.0858 0.1483 0.1483 0.1483 0.3080 0.1483
Table A.2. Distance among provinces of Iran (parameter of D [i ,j ]). Provinces East Azarbaijan West Azarbaij an Ardabil Isfa han Alborz Ilam Bushehr Tehran Chaharmahal an d Bakhtiar i S outh Khorasan Khorasan- e-Razavi K horasan-e-Shomal i Khu zestan Zanjan Se mnan Sis tan and Baluchestan East Azarbaij an 0 308 219 1,038 589 772 1,560 599 1, 142 1,912 1,493 1,321 1,075 280 835 2,166 West Azarbaij an 308 0 527 1,074 721 766 1,549 907 1, 178 2,220 1,802 1,620 1,064 588 1,143 2,264 Ardabil 219 527 0 1,030 545 975 1610 591 1, 134 1,814 1,333 1,080 1,305 377 828 2,154 Isfahan 1038 1,074 1,030 0 452 678 580 439 104 1,173 1,222 1,152 745 757 675 1,190 Alborz 589 721 545 452 0 691 1,082 48 583 1174 954 790 842 293 271 1,516 Ilam 772 766 975 678 691 0 932 710 719 1,788 1,604 1,423 447 598 946 1,868 Bushehr 1,560 1,549 1610 580 1, 082 932 0 1,228 684 1,599 1,648 1,941 485 1,338 1,464 1,404 Tehran 599 907 591 439 48 710 1,228 0 543 1,313 894 713 874 319 236 1,567 Chaharmahal and Bakhtiari 1,142 1,178 1,134 104 583 719 684 543 0 1,277 1,326 1,256 849 862 779 1,294 South K horasan 1,912 2,220 1,814 1,173 1, 174 1,788 1,599 1,313 1, 277 0 481 734 1,918 1623 1,139 470 Khorasan-e-Razavi 1,493 1,802 1,333 1,222 954 1,604 1,648 894 1, 326 481 0 253 1,768 1,213 658 951 Khorasan-e-Shomali 1,321 1,620 1,080 1,152 790 1,423 1,941 713 1, 256 734 253 0 1,587 1,032 543 1,204 Khuzestan 1,075 1,064 1,305 745 842 447 485 874 849 1918 1,768 1,587 0 967 1,110 1,759 Zanjan 280 588 377 757 293 598 1,338 319 862 1,623 1,213 1,032 967 0 555 1,886 Semnan 835 1,143 828 675 271 946 1,464 236 779 1,139 658 543 1,110 555 0 1,609 Sistan and Baluchestan 2,166 2,264 2,154 1,190 1, 516 1,868 1,404 1,567 1, 294 470 951 1,204 1,759 1,886 1,609 0 Fars 1,523 1,559 1,515 485 939 1,100 304 924 589 1,325 1,374 1,637 659 1,243 1,160 1,100 Qazvin 455 763 451 480 110 617 1060 150 584 1463 1,044 863 882 175 386 1,717 Qom 731 1,039 723 279 189 684 876 132 367 1,445 1,026 845 715 451 368 1,375 Kordestan 52 446 655 627 505 320 1,108 501 732 1,814 1,395 1,214 623 278 737 1,818 Kerman 1,637 1,735 1,629 661 1, 026 1,339 875 1,038 765 999 889 1,142 1,230 1,357 1,274 529 Kermanshah 588 582 791 653 519 184 972 526 731 1,800 1,420 1,239 487 414 762 1,817 Kohgeluyeh and Boyer-Ahmad 1,337 1,373 1,329 299 797 977 281 738 229 1,405 1,454 1,451 433 1,057 974 1,274 Golestan 996 1,304 764 836 454 1,107 1,625 397 940 1,050 569 316 1,271 716 377 1,520 Gilan 485 793 266 764 284 774 1,524 325 868 1,548 1,067 814 1,039 348 561 1,892 Lorestan 879 783 930 370 505 308 860 499 474 1,543 1,393 1,212 375 592 735 1,560 Mazandara n 866 1,174 634 706 320 977 1,495 267 810 1,180 699 446 1,141 586 205 1,650 Markazi 785 786 843 288 298 514 868 293 392 1,606 1,187 1,006 581 505 529 1,478 Hormozg an 1,933 2026 1,925 975 1, 317 1,729 927 1,334 1, 061 1,213 1,374 1,627 1,278 1,653 1,570 743 Hamadan 609 610 667 464 337 373 1044 337 568 1,637 1,231 1,050 638 329 573 1,654 Yazd 1,276 1,374 1268 300 664 978 726 677 404 873 922 1,390 1,081 996 913 890
Table A.2. (Continued ) Provinces Fars Qazvin Qom Kordestan Kerman Kermanshah Kohgeluyeh and Boyer-Ahmad Golestan Gilan Lorestan Mazandaran Markazi Hormozgan Hamadan Yazd East Azarbaijan 1,523 455 731 52 1,637 588 1,337 996 485 879 866 785 1,933 609 1,276 West Azarbaijan 1,559 763 1,039 446 1,735 582 1,373 1,304 793 783 1,174 786 2,026 610 1,374 Ardabil 1,515 451 723 655 1,629 791 1,329 764 266 930 634 843 1,925 667 1,268 Isfahan 485 480 279 627 661 653 299 836 764 370 706 288 975 464 300 Alborz 939 110 189 505 1,026 519 797 454 284 505 320 298 1,317 337 664 Ilam 1,100 617 684 320 1,339 184 977 1,107 774 308 977 514 1,729 373 978 Bushehr 304 1,060 876 1,108 875 972 281 1,625 1,524 860 1,495 868 927 1,044 726 Tehran 924 150 132 501 1,038 526 738 397 325 499 267 293 1,334 337 677 Chaharmahal and Bakhtiari 589 584 367 732 765 731 229 940 868 474 810 392 1,061 568 404 South Khorasan 1,325 1,463 1,445 1814 999 1,800 1,405 1,050 1,548 1,543 1,180 1606 1,213 1,637 873 Khorasan-e-Raz avi 1,374 1,044 1,026 1,395 889 1,420 1,454 569 1,067 1,393 699 1,187 1,374 1,231 922 Khorasan-e-Shoma li 1,637 863 845 1,214 1,142 1,239 1,451 316 814 1,212 446 1,006 1,627 1,050 1,390 Khuzestan 659 882 715 623 1,230 487 433 1,271 1,039 375 1,141 581 1,278 638 1,081 Zanjan 1,243 175 451 278 1,357 414 1,057 716 348 592 586 505 1,653 329 996 Semnan 1,160 386 368 737 1,274 762 974 377 561 735 205 529 1,570 573 913 Sistan and Baluchestan 1,100 1,717 1,375 1,818 529 1,817 1,274 1,520 1,892 1,560 1,650 1,478 743 1,654 890 Fars 0 965 764 1,113 571 1,112 1,74 1,321 1,249 855 1,191 773 619 949 425 Qazvin 965 0 282 453 1,172 433 779 547 185 507 417 303 1,455 244 780 Qom 764 282 0 474 846 499 595 529 457 340 399 134 1,142 289 485 Kordestan 1113 453 474 0 1,289 136 927 898 565 427 768 340 1,585 164 928 Kerman 571 1,172 846 1,289 0 1,288 745 1,435 1,363 1,031 1,305 949 485 1,125 361 Kermanshah 1,112 433 499 136 1,288 0 952 923 590 320 793 365 1,769 189 953 Kohgeluyeh and Boyer-Ahmad 174 779 595 927 745 952 0 1,135 739 699 1,005 587 756 763 532 Golestan 1,321 547 529 898 1,435 923 1,135 0 498 896 130 690 1,731 734 1,074 Gilan 1,249 185 457 565 1,363 590 739 498 0 664 368 577 1,659 401 1,002 Lorestan 855 507 340 427 1,031 320 699 896 664 0 766 206 1,327 263 670 Mazandaran 1,191 417 399 768 1,305 793 1,005 130 368 766 0 560 1,601 604 944 Markazi 773 303 134 340 949 365 587 690 577 206 560 0 1,245 176 588 Hormozgan 619 1,455 1,142 1,585 485 1,769 756 1,731 1,659 1,327 1,601 1,245 0 1,421 657 Hamadan 949 244 289 164 1,125 189 763 734 401 263 604 176 1,421 0 734 Yazd 425 780 485 928 361 953 532 1,074 1,002 670 944 588 657 734 0
References
1. A. Greenberg, J. R. Hamilton, N. Jain, S. Kandula, C. Kim, Lahiri et al., VL2: A scalable and °exible data center network, in ACM SIGCOMM Computer Communication Review (2009), pp. 51–62.
2. K. Kant, Data center evolution: A tutorial on state of the art, issues, and challenges, Computer Network 53 (2009) 2939–2965.
3. W. Fang, X. Liang, S. Li, L. Chiaraviglio and N. Xiong, VMPlanner: Optimizing virtual machine placement and tra±c °ow routing to reduce network power costs in cloud data centers, Computer Network 57 (2013) 179–196.
4. M. Al-Fares, A. Loukissas and A. Vahdat, A scalable, commodity data center network architecture, ACM SIGCOMM Computer Communication Review 38 (2008) 63–74. 5. K. J. Knapp, G. D. Denney and M. E. Barner, Key issues in data center security: An
investigation of government audit reports, Government Information Quarterly 28 (2011) 533–541.
6. X. Wang, Z. Du, Y. Chen, S. Li, D. Lan, G. Wang and Y. Chen, An autonomic provi-sioning framework for outsourcing data center based on virtual appliances, Cluster Computing 11 (2008) 229–245.
7. S.-J. F. Chang, S. H. Patel and J. M. Withers, An optimization model to determine data center locations for the army enterprise, in Military Communications Conference, MILCOM 2007, IEEE (2007), pp. 1–8.
8. M. I. M. Aliev and L. H. Kerimova, Optimal location of internet data centers taking into account the risks, Automatic Control and Computer Sciences 43 (2010) 309–316. 9. F. Yang and L. X. Ye, Method of locating data center based on Delphi, in 2011 Second
Int. Conf. Innovations in Bio-Inspired Computing and Applications (2011).
10. I. Goiri, K. Le, J. Guitart, J. Torres and R. Bianchini, Intelligent placement of data-centers for internet services, in 2011 31st International Conference Distributed Com-puting Systems (2011), pp. 131–142.
11. F. Larumbe and B. Sanso, Optimal location of data centers and software components in cloud computing network design, Presented at the 2012 12th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (ccgrid2012).
12. M. T. Covas, C. A. Silva and L. C. Dias, Multicriteria decision analysis for sustainable data centers location, International Transactions in Operational Research 20 (2013) 269–299.
13. M. T. Covas, C. A. Silva and L. C. Dias, On locating sustainable Data Centers in Portugal: Problem structuring and GIS-based analysis, Sustainable Computing: Infor-matics and Systems 3 (2013) 27–35.
14. T. U. Daim, A. Bhatla and M. J. I. J. O. S. E. Mansour, Site selection for a data centre– a multi-criteria decision-making model, International Journal of Sustainable Engineering 6 (2013) 10–22.
15. M. Klinkowski, K. Walkowiak and R. Goścień, Optimization algorithms for data center location problem in elastic optical networks, in Transparent Optical Networks (ICTON), 15th Int. Conf. (2013), pp. 1–5.
16. H.-A. Ouni¯, M. Ouhimmou, M. Paquet and J. J. I.- Montecinos, Datacenter location for Internet services in the province of Quebec, IFAC-Papers Online 48(2015) 2266–2272. 17. V. Depoorter, E. Oro and J. J. A. E. Salom, The location as an energy e±ciency and
renewable energy supply measure for data centers in Europe, Applied Energy 140 (2015) 338–349.
18. I. S. Mustafa, N. M. Din, A. Ismail, C. R. Omar and N. Khalid, Site suitability analysis for landslide monitoring base station using GIS-based multicriteria evaluation technique,
in 2015 5th National Symposium Information Technology: Towards New Smart World (NSITNSW), IEEE (2015), pp. 1–6.
19. J. Liu, P. Li, T. Shi and X. Ma, Optimal site selection of China railway data centers by the PSO algorithm, in 2016 12th World Congress on Intelligent Control and Automation (WCICA), IEEE (2016), pp. 251–257.
20. H. Zhang, G. Kou and Y. Peng, Soft consensus cost models for group decision making and economic interpretations, European Journal of Operational Research 277 (2019) 964–980.
21. G. Tian, H. Zhang, Y. Feng, D. Wang, Y. Peng and H. Jia, Green decoration materials selection under interior environment characteristics: A grey-correlation based hybrid MCDM method, Renewable and Sustainable Energy Reviews 81 (2018) 682–692. 22. T. Mukherjee, A. Banerjee, G. Varsamopoulos, S. K. S. Gupta and S. Rungta,
Spatio-temporal thermal-aware job scheduling to minimize energy consumption in virtualized heterogeneous data centers, Computer Networks 53 (2009) 2888–2904.
23. G. Tian and Y. Liu, Energy-e±cient models of sustainable location for a vehicle inspec-tion stainspec-tion with emission constraints, IEEE Transacinspec-tions on Automainspec-tion Science and Engineering 12 (2014) 238–243.
24. G. Tian, H. Zhang, Y. Feng, H. Jia, C. Zhang, Z. Jiang, Z. Li and P. Li, Operation patterns analysis of automotive components remanufacturing industry development in China, Journal of Cleaner Production 164 (2017) 1363–1375.
25. C. A. Mendez, J. Cerda, I. E. Grossmann, I. Harjunkoski and M. Fahl, State-of-the-art review of optimization methods for short-term scheduling of batch processes, Computers & Chemical Engineering 30 (2006) 913–946.
26. J. Rezaei, Best-worst multi-criteria decision-making method, Omega 53 (2015) 49–57. 27. J. Rezaei, Best-worst multi-criteria decision-making method: Some properties and a linear
model, Omega 64 (2016) 126–130.
28. M. Brunelli and J. Rezaei, A multiplicative best–worst method for multi-criteria decision making, Operations Research Letters 47 (2019) 12–15.
29. G. van de Kaa, D. Scholten, J. Rezaei and C. Milchram, The battle between battery and fuel cell powered electric vehicles: A BWM approach, Energies 10 (2017) 1707. 30. J. Rezaei, A concentration ratio for non-linear best worst method, International Journal
of Information Technology and Decision Making (2020).
31. D. Pamučar, L. Gigović, Z. Bajić and M. Janošević, Location selection for wind farms using GIS multi-criteria hybrid model: An approach based on fuzzy and rough numbers, Sustainability 9 (2017) 1–23.
32. S. Kheybari, M. Kazemi and J. Rezaei, Bioethanol facility location selection using best-worst method, Applied Energy 242 (2019) 612–623.
33. Ž. Stević, D. Pamučar, M. Subotić, J. Antuchevičiene and E. K. Zavadskas, The location selection for roundabout construction using Rough BWM-Rough WASPAS approach based on a new Rough Hamy aggregator, Sustainability 10 (2018) 2817.
34. S. Kheybari, F. M. Rezaie and J. Rezaei, Measuring the importance of decision-making criteria in biofuel production technology selection, IEEE Transactions on Engineering Management (2019) 1–15.
35. H. Amoozad Mahdiraji, S. Arzaghi, G. Stauskis and E. Zavadskas, A hybrid fuzzy BWM-COPRAS method for analyzing key factors of sustainable architecture, Sustainability 10 (2018) 1626.
36. H.-C. Liu, M.-Y. Quan, Z. Li and Z.-L. Wang, A new integrated MCDM model for sustainable supplier selection under interval-valued intuitionistic uncertain linguistic environment, Information Sciences 486 (2019) 254–270.
37. G. van de Kaa, L. Kamp and J. Rezaei, Selection of biomass thermochemical conversion technology in the Netherlands: A best worst method approach, Journal of Cleaner Production 166 (2017) 32–39.
38. Y. Yang, J. Ren, H. S. Solgaard, D. Xu and T. T. Nguyen, Using multi-criteria analysis to prioritize renewable energy home heating technologies, Sustainable Energy Technologies and Assessments 29 (2018) 36–43.
39. X. F. Ding and H. C. Liu, A new approach for emergency decision-making based on zero-sum game with Pythagorean fuzzy uncertain linguistic variables, International Journal of Intelligent Systems 34 (2019) 1667–1684.
40. H.-C. Liu, Y.-P. Hu, J.-J. Wang and M. Sun, Failure mode and e®ects analysis using two-dimensional uncertain linguistic variables and alternative queuing method, IEEE Transactions on Reliability 68 (2018) 554–565.
41. J. Huang, X.-Y. You, H.-C. Liu and S.-L. Si, New approach for quality function deployment based on proportional hesitant fuzzy linguistic term sets and prospect theory, International Journal of Production Research 57 (2019) 1283–1299.
42. W. N. K. W. Ahmad, J. Rezaei, S. Sadaghiani, and L. A. Tavasszy, Evaluation of the external forces a®ecting the sustainability of oil and gas supply chain using best worst method, Journal of Cleaner Production 153 (2017) 242–252.
43. K. W. Ahmad, M. de Brito, J. Rezaei and L. A. Tavasszy, An integrative framework for sustainable supply chain management practices in the oil and gas industry, Journal of Environmental Planning and Management 60 (2017) 577–601.
44. W. N. K. W. Ahmad, J. Rezaei, L. A. Tavasszy and M. de Brito, Commitment to and preparedness for sustainable supply chain management in the oil and gas industry, Journal of Environmental Management 180 (2016) 202–213.
45. R. Keeney and H. Rai®a, Decisions with Multiple Consequences: Preferences and Value Tradeo®s (John Wiley & Sons, New York, NY, 1976).
46. J. Rezaei, Piecewise linear value functions for multi-criteria decision-making, Expert Systems with Applications 98 (2018) 43–56.
47. S. Kheybari, F. M. Rezaie, S. A. Naji and F. Naja¯, Evaluation of energy production technologies from biomass using analytical hierarchy process: The case of Iran, Journal of Cleaner Production 232 (2019) 257–265.
48. P. Wanke, C. P. Barros and A. Emrouznejad, Assessing productive e±ciency of banks using integrated Fuzzy-DEA and bootstrapping: A case of Mozambican banks, European Journal of Operational Research 249 (2016) 378–389.
49. K. Pfe®er, Integrating spatio-temporal environmental models for planning ski runs, Phd diss (2003).
50. A. Lusardi and O. S. Mitchell, The economic importance of ¯nancial literacy: Theory and evidence, Journal of Economic Literature 52 (2014) 5–44.
51. J. Stiglitz, A. Sen and J. Fitoussi, The measurement of economic performance and social progress revisited: Re°ections overview, Commission on the Measurement of Economic Performance Social Progress, Paris (2009).
52. N. Habibi, The Iranian economy in the shadow of economic sanctions, Middle East Brief 31 (2008) 1–8.
53. W. Kang, R. A. Ratti and J. Vespignani, The impact of global uncertainty on the global economy, and large developed and developing economies, CAMA Working Paper (2017). 54. A. Kia, Fiscal sustainability in emerging countries: Evidence from Iran and Turkey,
Journal of Policy Modeling 30 (2008) 957–972.
55. A. P. Groh and M. Wich, Emerging economies' attraction of foreign direct investment, Emerging Markets Review 139 (2012) 210–229.
56. A. Ehteshami and R. A. Hinnebusch, Syria and Iran: Middle Powers in a Penetrated Regional System (Routledge, 2002), pp. 37–66.
57. M. Ahmad, T. Ali and Z. J. J. O. S. Shah, The US factor in Pakistan–Iran relations: New dimensions, Journal of Political Studies 24 (2017).
58. G. J. L. Lazaroiu, Employee motivation and job performance, in Linguistic and Philo-sophical Investigations (2015), pp. 97–102.
59. T. Brunschwiler, B. Smith, E. Ruetsche and B. J. I. J. O. R. Michel, Development, toward zero-emission data centers through direct reuse of thermal energy, IBM Journal of Research and Development 53 (2009) 1–11.
60. E. Pakbaznia and M. Pedram, Minimizing data center cooling and server power costs, in Proc. 2009 ACM/IEEE International Sympsium on Low Power Electronics and Design (2009), pp. 145–150.
